Video Preview
🎧 Podcast episode for paid subscribers only. Also available on Apple Podcasts and Spotify.
To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast).
Table of Contents
The claim, and the four words doing all the work in it
What a polygenic score for intelligence is actually measuring
The sibling problem, and why within-family validation is the whole ballgame
Running the five-embryo math the way a clinic would run it
Selection versus editing, which are not the same product at all
Pleiotropy, ancestry, and the things you moved without meaning to
The regulatory map, which is mostly white space
Unit economics, the IVF funnel, and who actually pays
What the customer is buying, which is not IQ points
How this probably goes from here
Abstract
A consumer genomics company has shipped a set of trait models it calls Vitruvian, marketed with the line that its intelligence model can optimize embryo DNA for 14 IQ points, close to a full standard deviation. The supporting material is more specific and more modest: training on 1M+ people, validation across 40,000+ sibling pairs, 7M+ common variants scored with a Bayesian method using 96 functional annotations, and an expected spread of roughly 10.8 IQ points between the highest and lowest scoring embryo in a set of five. That spread number is a range statistic, not a gain, and the gain against an average embryo is roughly half of it. Older peer-reviewed modeling of the same problem put the realistic expected gain at about 2.5 points when picking the best of 10. The gap between 2.5 and 10.8, and between a roughly 5‑point gain and a 14‑point marketing line, is where the entire argument lives. This essay works through the statistical machinery (heritability vs. predictability, between-family vs. within-family R², range vs. expected maximum), the clinical funnel (median euploid blastocyst counts of 2 to 4, not 10), the pleiotropy and ancestry-portability problems, the regulatory vacuum (a CLIA lab test that no agency clinically validates, an FDA appropriations rider that only blocks editing, and state embryo law as the real tail risk), and the business model underneath it all (roughly 413,776 US ART cycles a year, around 2.3% of births, $5,999 cash-pay add-on, no payer coverage). Short version: the biology is real, the scores are better than they were, the marketing number is a best case dressed as a product spec, and the thing being sold is not points.
